Can you predict the future? Most people would say absolutely not, certainly not in the sense of making highly accurate forecasts about what or when something will happen. But not everyone agrees. Dirk Helberg from the Federal Institute of Technology in Zurich has suggested a scientifically based way of putting all the world’s data into a single super computer, which then predicts what the future will look like. Astonishingly, perhaps, the debt-ridden European Union is considering giving Helberg one billion euros to build such a machine.
The idea, in a nutshell, is an extension of the Big Data idea. Take every bit of available data across economics, government, cultural trends, energy, agriculture, health, technological developments, together with data linked to climate and weather to create a Living Earth Simulator or FuturICT Knowledge Accelerator as it’s called. Why would such an idea work? Because Mr Helberg says he’s done it before – kind of. Some time ago Helberg built a system to model highway traffic, which showed that reducing the distance between vehicles resulted in an end to stop-go delays. The only problem is that for this to happen you have to reduce the space between moving vehicles to such a small distance that you need self-driving cars to make it work. Moreover, a highway is less complex that the whole earth with seven billion highly emotional, at times irrational and occasionally anarchic human inhabitants.
As Gary Kind, director of the Institute for Quantitative Social Science at Harvard says, agent-based modelling only works when you are dealing with a narrow set of circumstances. For instance, how do you model the emotional impact of the death of a world leader, the arrival of UFOs, $200 oil or 9/11? The timing of such events is surely impossible to predict and there are always complex feedback loops. Moreover, we do not have a rigorous model for human behaviour and humans, ultimately, are at the heart of what Helberg wants to model.
It’s true that with enough data one can build sophisticated models, even if we do not understand the laws governing certain types of behaviour. We can look for patterns and anomalies and these may be used to predict outcomes or create policy. We are also on the cusp of a world with unparalleled volumes of data and analytical sophistication.
Nevertheless, predicting precisely where and when a financial market will collapse or a global pandemic will start is vastly more complex than predicting highway traffic flows. We can’t even agree on what will happen in financial markets tomorrow let alone next year or in five years time. In my opinion Helberg is falling face first into a number of huge traps. The first is that his idea is dependent upon extrapolation from historical data. He is assuming that everything has a rational explanation and that everything can be measured and modelled. But I’d argue that things often happen for no observable reason (dumb luck) or that even when there’s a clear reason the chain of events that follow cannot be accurately predicted. Take 17 December 2010 for instance. How can you create a machine that predicts that a street vendor in a small Tunisian town will on this day set himself on fire and that his protest will create a string of popular revolutions across the Middle East? You might observe that the conditions are right for such an event to take place, eventually, but you can’t say when or where or and so. It’s like saying that an area of land contains dry kindling and that one day it will be set on fire, but you cannot say when the fire will break out or how far – and how fast – the fire will subsequently spread.
Apart from complexity and chaos there’s also the issue of humans not understanding outcomes. What if, for example, the machine says that to prevent a health pandemic we need to kill a democratically elected politician? It would make no sense to us so should we do it? We would have knowledge of the problem, perhaps, but we would not be able to comprehend or understand the machine’s solution. Moreover, if the machine made several suggestions, how would we decide which course of action to take? Another issue is that of the self-fulfilling prophecy. If a trusted model says something will happen then a prediction can impact the situation being modelling. This happened recently in the UK when the government said that a fuel tanker strike might give rise to fuel shortages. Everyone took this prediction at face value and went into panic mode, thereby creating a fuel shortage. Finally, a centralised system of data collection and analysis is a very old fashioned idea. Instead, why not create clouds of open data via the Internet and give access to everything to everyone. An open data format would encourage participation, collaboration and fruitful disagreement. This would cost far less and work far better.
BTW, there’s a computer called Nautilus in the US that reads the news and works out what’s next…kind of. Similar sort of thing.